Erroneous Consonance. How inaccurate beliefs about physician opinion influence COVID-19 vaccine hesitancy

Abstract

Throughout the COVID-19 pandemic, researchers have studied how Americans' attitudes toward health experts influence their health behaviors and policy opinions. Fewer, however, consider the potential gap between individual and expert opinion about COVID-19, and how that might shape health attitudes and behavior. This omission is notable, as discrepancies between individual and expert opinion could help explain why some Americans fail to take action to protect themselves and others from the virus. In novel demographically representative surveys of the US adult population (N = 5,482) and primary care physician subpopulations (PCPs; N = 625), we contrast the relationship between: (1) Americans’ and (2) PCPs' preferences regarding who ought to be responsible for taking action to combat the spread of COVID-19, as well as (3) Americans' perceptions of PCP preferences (“PCP meta-opinion”). In the aggregate, we find that Americans are far less likely than PCPs to see a role for both private and state actors in taking action to combat COVID-19. Interestingly, though, this disjuncture is not reflected in individual-level PCP meta-opinion; as most Americans think that PCPs share their views on state and private intervention (τ_b = 0.44–0.49). However, this consonance is often erroneous, which we show can have problematic health consequences. Multivariate models suggest that Americans who both see little place for individual responsibility in taking action to stop viral spread and who think that PCPs share those views are significantly less likely to vaccinate against COVID-19. We conclude by discussing the public health benefits of efforts to bring public opinion in line with expert opinion.

1. Introduction

The availability of safe and effective vaccines has played a key role in curtailing the COVID-19 pandemic’s spread and severity. Nevertheless, many Americans are reluctant to vaccinate against COVID-19 ([9], [13]). Correspondingly, substantial scholarly interest has been devoted to studying how public opinion regarding individuals’ responsibility to stop the spread of COVID-19 relates to vaccine hesitancy (e.g., [29], [4]).

There are (at least) two pathways by which public opinion about who ought to be responsible for taking action to stop the spread of the COVID-19 might influence public opinion. The first concerns opinion itself, or what some call “first order” opinion. For example, Americans who see less of a role for individual action in combating the effects of the COVID-19 pandemic may be less likely to take protective action (e.g., vaccine uptake; physical distancing) against disease spread.

A wealth of research at the intersection of social science and behavioral medicine documents the effects of perceived individual responsibility on Americans’ vaccine attitudes and behavior. Those who hold strong prosocial motivations and recognize the societal benefits of vaccination are more likely to hold strong views of collective responsibility in population health and disease prevention through vaccination [22], [39]. However, Americans who prioritize the protection of their civil liberties or view vaccination as an individual – rather than collective – responsibility tend to exhibit higher levels of vaccine hesitancy [2], [4], [32]. The same holds for COVID-19 vaccination attitudes, in particular. Americans who believe in societal responsibility for vaccination are more likely to have received or intend to receive a COVID-19 vaccine than those who view vaccination as a purely individual choice or responsibility [1], [3], [20].

A second way that public opinion might influence health outcomes is via opinion about opinions; what some call “second order” opinion or “meta-opinion” [28]. According to this view, Americans’ views about what others think -- including the opinions of medical experts -- might influence their own health attitudes and behavior. When Americans view experts’ opinions as similar to their own, despite gulfs in the groups’ respective preferences, this is known as a “false consensus effect.” [35], [26]. False consensus effects are problematic, as they can be used as the basis to (incorrectly) rationalize one’s own attitudes and behaviors [28]. In the domain of public health, for example, Americans who (incorrectly) think that physicians – a group that the public generally trusts to provide them with accurate health information [15] – see little role for individual action in stopping the spread of COVID-19 may use this information to justify their own inaction to limit virus transmission through vaccination. While understudied in health applications, past meta-opinion research has found that underestimation of public anthropogenic climate change acceptance is associated with opposition to mitigatory climate policy action [28] [36], [17].

Less clear from past work, however, are how false consensus effects -- in reference to medical expert opinion, and second-order opinion more generally -- might influence vaccine uptake during the COVID-19 pandemic. Given the high levels of trust that most Americans place in their own physicians ([15]), false consensus could exert a powerful influence on vaccine-related attitudes and behavior.

Past work has studied the effect of perceived social norms on vaccine uptake and preventative behaviors (see: [6], [7] for extensive reviews). For example, the tendency to avoid physically distancing oneself from close kin might encourage non-compliance with mask-wearing and social distancing behavior in the COVID-19 pandemic [31]. But, as social norms are necessarily opinions about non-expert opinions, this important research has somewhat limited applicability to our own research on expert-related effects.

Moreover, the link between more-general descriptive norms and vaccine uptake may depend on the referent group seen as promoting those views. Highlighting medical expert agreement about childhood vaccine safety has been shown to increase pro-vaccine attitudes by raising the extent to which the public is aware of expert consensus (e.g., [38]. However, whereas past research on peer social groups has observed high levels of intended COVID-19 vaccine uptake despite low perceived uptake [18], family members and close friends have been shown to exert a stronger influence on vaccine attitudes and behavior [34].

Taken together, we believe that this paper is the first to assess the relationship between Americans’ meta-opinion about expert beliefs and the role that individual Americans ought to take in combating COVID-19. Correspondingly, the prevalence of false consensus, at both the individual (RQ1) and mass level (RQ2), with respect to expert opinion on this score has yet to be determined. Because meta-opinion can theoretically play a powerful role in shaping what people view as acceptable health behavior, we hypothesize that people who underestimate the importance that medical experts place on taking individual action to combat viral spread should be less likely to vaccinate against COVID-19 (H1).

In this piece, we study Americans’ opinions, primary care physicians’ (PCP) opinions, and Americans’ opinions about primary care physicians’ (PCP) opinions (“PCP meta-opinion”), regarding the role that different private (i.e., individuals, businesses) and state actors (i.e., the federal government, state and local governments) ought to play in taking action to stop the spread of COVID-19. We find that Americans tend to believe PCP opinions are closer to their own opinions, when in fact PCP opinions differ substantially. Throughout this piece, we refer to this false consensus about medical experts’ COVID-19 opinion as erroneous consonance.

Specifically, we quantify erroneous consonance in the medical community’s COVID-19 opinions in two ways. First, we assess the consonance between (a) aggregated measures of public opinion, (b) PCP meta-opinion, and (c) actual PCP opinion in demographically representative surveys of the US adult population (quantities a-b) and PCP populations (quantity c). We then disaggregate these broader trends by assessing the relationship between individual-level opinion about these issues and perceived PCP opinion. We use these later micro-level measures to study the effects of erroneous consonance on behavioral self-reports of COVID-19 vaccine uptake.

2. Data & measures

Our study features two original and demographically representative surveys collected over the course of the COVID-19 pandemic. First, we collected a demographically representative survey of 5,482 US adults collected via Lucid Theorem from June 24–28, 2022. Lucid uses quota sampling procedures to collect responses from their large online panel of potential participants, targeting representativeness to the US population on the basis of age, race/ethnicity, level of education, and income. A comparison of our sample of from the general public to population benchmarks from the US Census in the supplementary materials demonstrates that our sample closely mirrors population benchmarks. Importantly, prior research demonstrates that survey data from Lucid (a) closely resembles demographic benchmarks derived from nationally representative benchmarks, (b) produces estimates of both political and health attitudes/behavior similar to those derived from probability samples of US adults, (c) has been employed widely in the study of vaccine hesitancy, and (d) has been shown to replicate well-studied experimental treatment effects both prior to and throughout the pandemic ([12], [9], [33], [30], [21].

Second, our study includes an original survey of primary care physicians in the United States. This survey was administered to a national sample of 737 physicians using the survey research firm Dynata from May 14–25, 2021. Dynata has extensive experience collecting data from specialized professional subsamples and is widely used in social science research [24], [11], [25], [23], [10]). Dynata invited primary care physicians from their large online opt-in panel – identified via an initial inventory survey – to participate in our study. Of the 737 physicians Dynata recruited, 625 self-identified as primary care physicians (PCPs) – working in either family medicine, internal medicine, or general practice – and were included in our sample. Notably, prior published research using this dataset [10], [16], demonstrates that our sample closely approximates population benchmarks for primary care physicians in the United States.

The primary outcome variables used to assess the prevalence of erroneous consonance are ordinal measures of the extent to which Americans and physicians think that each of five different groups – i.e., (1) local, (2) federal, and (3) state governments, as well as (4) individuals and (5) businesses – ought to play a role in “managing the COVID-19 pandemic.” Respondents could indicate that each group should be “very responsible,” “somewhat responsible,” “not very responsible,” or “not at all responsible” for doing so. These questions were identically phrased in both surveys. We rely on this set of outcomes to measure erroneous consonance because they allow us to assess the extent to which the general public and physicians believe that individuals, as opposed to other actors, have a responsibility to play in managing the pandemic; with the primary individual behaviors focusing on the adoption of preventive health behaviors like masking and vaccinating.

Additionally, we assess the consequences of erroneous consonance for COVID-19 vaccine uptake by first asking respondents whether or not they had “personally received the COVID-19 vaccine, or not.” Respondents could indicate in the affirmative that they “got a one-dose vaccine,” “got the first dose of a two-dose vaccine,” “got both doses of a two-dose vaccine” or “have not been vaccinated.” Individuals selecting the first and third options were classified as being fully vaccinated (taking on a value of 1, and 0 otherwise). Additionally, we asked respondents whether or not they had “personally received at least one booster dose of the COVID-19 vaccine,” to which respondents could answer “Yes” or “No” (dichotomized such that affirmative responses take on a value of 1, 0 otherwise).

All multivariate models employed to assess the consequences of erroneous consonance on vaccine uptake control for a wide range of social, psychological, and political factors that have been shown to be associated with COVID-19 vaccine hesitancy. These include an ordinal measures of anti-expert attitudes and ideological conservatism, and an intervalized measure of individualistic values [27], [9], [29], [14]. We also control for respondents’ racial and ethnic identity (dichotomous indicators of whether or not respondents self-identify as Black [Non-Hispanic] or Hispanic), a continuous income scale, and a dichotomous indicator of whether or not respondents have a college degree. All controls are scored to range from 0 to 1. More information about each measure can be found in the Supplementary Materials.

3. Results

3.1. Assessing the prevalence of erroneous consonance

We begin by assessing the prevalence of erroneous consonance in aggregated survey data (RQ1). Fig. 1 plots the percentage of survey respondents in the Lucid survey of the US adult population (denoted as S in the figure) and Dynata survey of US PCPs (denoted as A) who think that (in order) the federal government, state governments, local governments, individuals, and private businesses ought to be “very responsible” for “managing the COVID-19 pandemic.” It also presents an estimate of PCP meta-opinion (P); i.e., the percentage of Americans in the Lucid survey who perceive PCP’s as indicating that each group should be “very responsible” on this score.

Fig. 1.

The consonance Between Americans’ Opinion, PCP Opinion, and PCP meta-Opinion on State/Private Responsibility for Addressing COVID-19 (Macro-Level Survey Data). Note. Quantities S (individual attitudes) and P (PCP meta-opinion) are derived from a Lucid survey of US adults. Quantity A is derived from a Dynata survey of US PCPs. Please refer to the main text for additional information about these aggregated (macro-level) opinion estimates. Note that a character resembling the letter “B” appears in the “State” row, due to perfect overlap between S and P estimates.

Fig. 1 suggests that erroneous consonance is quite common. In all cases, US public opinion (S) is less likely than the opinion of PCPs (A) to suggest that each group should be “very responsible” for handling the COVID-19 pandemic. However, and indicative of the prevalence of erroneous consonance, Americans perceived physician opinions (P) to be quite similar to their own.

Erroneous consonance is particularly evident with respect to the role that individuals ought to play in addressing COVID-19. On that question, we document an 18 percentage point gap – the largest we observe in the series – between Americans’ preferences (53 % indicating that individuals ought to be “very responsible”) and those of PCPs (71 %). At the same time, however, PCP meta-opinion (52 %) is nearly identical to public opinion. These results suggest that while Americans and physicians differ substantially in their assessments of who ought to play a major role in addressing the COVID-19 pandemic, there is strong evidence of erroneous consonance between Americans’ preferences and Americans’ beliefs about where PCPs stand (i.e., PCP meta-opinion).

We also find strong evidence of erroneous consonance when turning to micro-level opinion data. Fig. 2 displays the cross-tabulation of individual survey respondent preferences and PCP meta-opinion in the Lucid survey of US adults. Each cell presents the proportion of respondents indicating each possible response combination, shaded according to their frequency.

Fig. 2.

The consonance Between Americans’ Opinion & PCP meta-Opinion on State/Private Responsibility for Addressing COVID-19 (Micro-Level Survey Data). Note. Figures listed in each cell correspond to the proportion of the sample selecting each opinion/meta-opinion combination. Letters listed in the figure correspond to each ordinal scale point; such that V = very responsible, S = somewhat responsible, NV = not very responsible, and NA = not at all responsible.

Darker shading (and higher proportions) listed on the “main diagonal” of each figure indicate higher degrees of (potentially erroneous) consonance between respondents’ own preferences and PCP meta-opinion. We also provide a more formal measure of consonance by listing τ_b coefficients – a measure of covariance appropriate for bivariate analyses between ordinal variables [8] – below each panel. These coefficients can be interpreted similarly to correlation coefficients, such that a score of 0 indicates no covariation between two ordinal measures, while scores of 1 and −1 (respectively) indicate perfect positive or negative covariation.

The results again provide strong evidence of erroneous consonance between survey respondents’ opinions and PCP meta-opinion. For example, when asked about the role that individuals ought to play in managing COVID-19 (bottom left), 63 % of respondents provide response combinations on the figure’s main diagonal; such that 40 % simultaneously think – and think that PCPs prefer – that individuals ought to be “very responsible” for handling COVID-19, 17 % think (and perceive) that this group should be “somewhat responsible,” and so on. Correspondingly, we detect a moderately high degree of positive covariation between the two quantities (τ_b = 0.47).

Given the large macro-level gaps between individual and PCP opinion documented in Fig. 1, and the high degrees of micro-level consonance between public opinion and PCP meta-opinion documented in Fig. 2, we again believe that these analyses provide strong evidence of erroneous consonance.

4. The effects of erroneous consonance on COVID-19 vaccine uptake

Having demonstrated the prevalence of erroneous consonance at both the macro and micro levels (RQ1), we next consider whether or not Americans who express these views are subsequently less likely to vaccinate against COVID-19. To do this, we first construct a series of micro-level dichotomous indicators that denote all possible combinations of (a) public opinion and (b) PCP meta-opinion pertaining to the role that individuals ought to play in taking action to combat COVID-19. These nominal indicators include cases where (1) public opinion and meta-opinion both indicate that individuals ought to be “very responsible,” for handling COVID-19 [denoted as “high/high” or “H/H”]; (2) public opinion and meta-opinion both indicate that individuals should not be “very responsible” (“low/low,” or “L/L”), and those cases where (3) public opinion (“L/H”) and (4) meta-opinion (“H/L”) disagree regarding whether or not individuals ought to be “very responsible” for addressing the pandemic.

Conceptually, we are primarily interested in the effects of falling into the second nominal category, as this denotes erroneous consonance among those who individuals ought to play a limited role in addressing COVID-19, and who (incorrectly) think that PCPs feel the same way. We study these effects by constructing logistic regression models that regress binary indicators of whether or not Americans are (a) fully vaccinated against COVID-19 and (b) have chosen to receive a supplemental “booster” vaccine on each of the nominal indicators listed above (with H/H serving as an analytical reference group) and all possible confounding factors described earlier (see: Measures).

For brevity, we provide the full regression model output in the Supplementary Materials. There, we find that – consistent with our expectations (H1) – the expression of erroneous consonance is both negatively and significantly (at the p < 0.05 level, two-tailed) associated with a decreased likelihood of being fully vaccinated (B = -0.68, p < 0.05) and receiving a booster vaccine (B = -0.61, p < 0.05).

Fig. 3 summarizes the substantive size of the results obtained from our logistic regression models by plotting the predicted probability that respondents who express erroneous consonance (and, for reference, all other combinations of public and meta-opinion) self-report being fully vaccinated against COVID-19 (Panel A) or having received a supplemental booster shot (Panel B). We calculate these quantities holding all other covariates at their sample means.

Fig. 3.

The Effects of “Erroneous consonance” on COVID-19 Vaccine Booster Uptake (Micro-Level Survey Data). Note. Predicted probabilities presented as bars, with 95% confidence intervals extending out from each one. Results are derived from multivariate logistic models described in text, and presented in full in the Supplemental Materials. Predicted probabilities hold all covariates at their sample means.

The results again underscore the potentially deleterious effects of erroneous consonance on vaccine uptake (H1). Compared to those for whom public and meta-opinion both suggest that individuals ought to be “very responsible” for responding to COVID-19 (“H/H”), respondents expressing erroneous consonance are less likely to be fully vaccinated (74 % for H/H vs 60 % for LL) or to have received a COVID-19 booster shot (59 % for H/H vs 46 % for LL). These differences in self-reported behavior are both substantively large – corresponding to a 14 percentage point difference for vaccine uptake, and a 13 percentage point difference for booster vaccination – and statistically significant (as indicated by the non-overlapping confidence intervals between the first and second bars).

Similarly, we find that those who express erroneous consonance are less likely to self-report vaccinating than those indicating all other possible opinion and meta-opinion combinations. However, we caveat that (at times) the differences in vaccine uptake between these two groups fails to attain two-tailed significance (as indicated by the overlapping confidence intervals).

Although we lacked a priori expectations about vaccine uptake behavior in these latter groups, the results provide further suggestive evidence of the potential role that PCP meta-opinion may play in shaping health behavior; i.e., because as respondents who believe that PCPs see a strong role for individual action in combating COVID-19 – despite disagreeing themselves (i.e., the “H/L” nominal category) – tend to vaccinate at a similar rate to those in the H/H category. Correspondingly, we see efforts to move beyond the study of erroneous consonance to further disentangle the effects of meta-opinion on vaccine uptake as a fruitful direction for future research.

Finally, we recognize that the results presented in Fig. 3 might prompt readers to ask why some Americans are more likely than others to express erroneous consonance. Readers may be particularly interested in whether or not those who express comparatively warmer feelings toward scientific experts are more likely to express consonance of any variety. Elevated levels of trust, in other words, might motivate people to assume that experts must, in-turn, share their own views on pandemic-related matters.

To this end, we explore the socio-political and demographic correlates of erroneous consonance by regressing the nominal indicators of PCP meta-opinion featured in Fig. 3 on all covariates used to produce that figure (see: Data & Methods). In analyses available in the Supplemental Materials, we find that individuals who hold more-positive views toward scientific experts are significantly more likely to express informed consonance (i.e., fall into the H/H group), but neither more nor less likely to express erroneous consonance (i.e., the L/L group). This offers some suggestive evidence that trust in scientific experts may facilitate (or, at the very least, not inhibit) Americans from assuming that medical experts share their views on pandemic-related issues.

Of course, we caveat that these analyses are preliminary, as our data are correlational in nature and therefore cannot definitively determine causality. For example, it could be the case that high levels of trust result from the perception of sharing opinions with the medical community. Nevertheless, we hope that these analyses can serve as a blueprint for future efforts to unpack the nature and origins of erroneous consonance (including the potential role that medical and vaccine knowledge may play in shaping these views), and to consider its application to other vaccine-related issues.

5. Conclusion & discussion

The results presented in this paper provide both macro and micro-level documentation of erroneous consonance between public opinion and PCP meta-opinion, particularly with respect to the role of individual responsibility in taking action to combat the spread of COVID-19. Troublingly, the results suggest that erroneous consonance could have important public health consequences. People who see less of a role for individual action in mitigating the spread of the pandemic, and who (incorrectly) think that PCPs share their opinions, are significantly less likely to self-report vaccinating against COVID-19.

Because PCPs are largely trusted on issues related to COVID-19, [37], [10] operate outside of government bureaucracy, and frequently interact with members of the public on an individual basis, we think that better public health messaging is needed to inform the public about PCP perceptions and scientific consensus regarding COVID-19 responsibility. This could include PCPs taking an active role in communicating with patients regarding scientific consensus on COVID-19 and the responsibility that individuals and other entities ought to play in curbing its spread.

That said, some (e.g., strong Republicans; [19]) may be less receptive to these communication efforts, particularly in light of the societal polarization around COVID-19 attitudes and behaviors in the US [9], [14]. Politicization may pose an obstacle to reducing erroneous consonance, and subsequently encouraging COVID-19 vaccine uptake. Individuals are more likely to accept information cues from trusted sources, such as political and media sources congruent with their own political affiliations. In the context of the COVID-19 pandemic, those on the right are more likely to accept information from conservative political and media sources that tend to be skeptical of collective sources of responsibility, and reject information associated with the political left; e.g., that responsibility lies with government and its ability to impose health interventions [5]. Accordingly, if people with certain political affiliations - particularly strong Republicans - incorrectly perceive medical consensus to be skewed toward their own opinions (thus helping justify their beliefs), then they may: 1) be resistant to new information that aims to provide accurate PCP attitudes because they would perceive it to align with the political left (thereby making it difficult to reduce erroneous consonance), and 2) lose trust in PCP sources that provide information that might otherwise cut against their previously-held beliefs about science and medicine.

Our study has several noteworthy limitations. First, we are unable to test the causal relationship between erroneous consonance and COVID-19 vaccine uptake. While our findings point to a clear correlation between the two, and while there are theoretical reasons to anticipate the hypothesized causal pathway, we urge future researchers to investigate the impact of erroneous consonance on vaccination and other health outcomes via longitudinal and/or experimental studies better suited to isolate causality. Second, the results of this study rely on online opt-in survey data, rather than nationally representative data. While we have made an effort to ensure both response quality and representativeness (see: Data & Measures), and believe that our study is well-suited to study associations between variables, we nevertheless caution readers from generalizing specific “point estimates” of opinion/meta-opinion to the general population. Next, while we believe that our questions about individual vs business and government responsibility are useful measures in our study of erroneous consonance, it is important to recognize that these questions do not focus on vaccination specifically. Future research would benefit from expanding on our research here to explore more vaccine-specific perceptions as well. Finally, we note that respondents may bring different considerations to mind when answering the survey questions used to measure erroneous consonance. We encourage future qualitative work on this subject to consider asking respondents open-ended survey questions pertaining to the different policy and/or health behavioral actions that might constitute, for example, what it means to “manage” the COVID-19 pandemic.

Open Data Statement

Data and syntax necessary to replicate all main text analyses are available at the Open Science Framework at https://osf.io/se4v8/?view_only = ee2023ae8e494bee89ae50d4dea9148b.

Funding Statement

The authors have no funding to report.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

The following are the Supplementary data to this article:

Data availability

Data & replication code are available at: https://osf.io/preprints/socarxiv/8hnxd/